MoWE-Audio: Multitask AudioLLMs with Mixture of Weak Encoders
This work addresses a bottleneck in AudioLLMs for processing diverse audio tasks, representing an incremental improvement.
The paper tackled the constrained capacity of pre-trained audio encoders in AudioLLMs for new tasks and datasets by proposing a mixture of weak encoders (MoWE) to enhance feature extraction without significantly increasing model size, resulting in improved multi-task performance.
The rapid advancements in large language models (LLMs) have significantly enhanced natural language processing capabilities, facilitating the development of AudioLLMs that process and understand speech and audio inputs alongside text. Existing AudioLLMs typically combine a pre-trained audio encoder with a pre-trained LLM, which are subsequently finetuned on specific audio tasks. However, the pre-trained audio encoder has constrained capacity to capture features for new tasks and datasets. To address this, we propose to incorporate mixtures of `weak' encoders (MoWE) into the AudioLLM framework. MoWE supplements a base encoder with a pool of relatively light weight encoders, selectively activated based on the audio input to enhance feature extraction without significantly increasing model size. Our empirical results demonstrate that MoWE effectively improves multi-task performance, broadening the applicability of AudioLLMs to more diverse audio tasks.